6.3 Monitoring Exchange 2003 Performance 475
Chapter 6
6.2.4 Conclusion
I view LoadSim and ESP as precise and focused tools that serve not only for
stressing a server, but also for providing you with insight data about a
benchmark environment and defining a baseline for your computing infra-
structure. Both of these tools can determine the capacity and vertical scal-
ability of the solution (i.e., how much it can handle). However, they do not
properly address the other components that are stressed by high activity and
are used by Exchange 2003, such as the Active Directory services. ADtest
and JetStress have proven their usefulness in testing respectively AD servers
and storage subsystems. I believe that the latter one is of crucial importance
to any scaled-up environment, and JetStress should be in the toolbox of any
person involved in the deployment and configuration of Exchange servers.
You will have to be very careful when properly analyzing the actual per-
formance data in the early stage of your deployment not to rely solely upon
your benchmark environment. Remember that simulation testing can bring
up interesting points but does not always represent an actual user workload:
an Outlook client can do many things (e.g., inbox processing rules, large
item count, additional mobility clients) that a Loadsim or JetStress environ-
ment cannot simulate.
6.3 Monitoring Exchange 2003 Performance
6.3.1 Introduction
Performance monitoring in Windows 2003 has evolved from the Perfor-
mance Monitor as we knew it with Windows NT 4 (and still available from
the Windows 2000 resource kit) into two distinct tools:
A system monitor, which displays current or saved performance
counters information;
Performance logs and alerts, which capture counter values and man-
age log files and alerts.
I value the system monitor GUI, but I still believe that post-processing
tools are often required to make a proper performance analysis and assess-
ment of a running Exchange server. Microsoft Excel can be quite handy for
implementing basic statistical and mathematical functions. Further on, I
476 6.3 Monitoring Exchange 2003 Performance
will discuss a methodology to process and turn out something meaningful
from several hundred megabytes of information.
The performance monitor in Windows 2003 comes with three basic fea-
tures:
Performance counters that capture into log files;
Trace information gathering (but what do you do with this trace
information?);
Alert generation based on counter threshold, which can be utilized for
triggering additional (perhaps more detailed) performance information.
I like the new monitor because it lets you concentrate on the data acqui-
sition task and does that well, by providing data in multiple formats and by
letting you schedule data acquisition (e.g., if you wanted to understand the
stress put on your server during an online database defragmentation during
off-peak hours). Granted, the performance counter selection dialog box
hasnt changed much, and the acquisition time interval is limited to one
second (just long enough). However, the monitor comes with interesting
log file management features, such as the ability to stop the acquisition after
a given amount of time or at a precise date/time. Sequence numbering of
logfile names is also handy.
In addition to Perfmon, some tools that were in previous versions of the
Windows Server resource kit are now included in the base operating system
(Windows XP or Windows Server 2003). These tools allow rapid capture of
performance data as well as refinement of performance data (TYPEPERF
and RELOG, which we discuss later in this section).
Monitoring Exchange 2003 can be both an easy and rather complex
task. It is easy because Microsoft, in both its operating systems and its appli-
cations, has always provided detailed information about the behavior of the
running applications. On the other hand, it can be complex; in the process
of searching for performance bottlenecks, you may be tempted to gather
many performance counters, and having too much data to process can
result in burying somewhat important information under irrelevant data.
Too much data kills the data.
You therefore need to find a compromise that allows you to get essential
data at a reasonable sampling interval. This way, you can easily report and
make decisions while not overlooking essential elements, and you can have

Get Microsoft® Exchange Server 2003 Scalability with SP1 and SP2 now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.